huawei-cloud-migration-vmware-ecs-private

Warn

Audited by Gen Agent Trust Hub on Sep 30, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADSCREDENTIALS_UNSAFE
Full Analysis
  • [COMMAND_EXECUTION]: Utility scripts such as batch_migrate.py, vpn_check.py, and generate_template.py execute local shell commands using subprocess.run or os.system with shell=True. These commands are used for self-daemonization, network connectivity validation, and missing dependency installation.
  • [DYNAMIC_EXECUTION]: The migration workflow dynamically generates and executes administrative shell scripts on remote servers via SSH connections. Notable examples include agent_state_check.sh for agent status monitoring and start_agent.sh for initiating the migration process.
  • [INDIRECT_PROMPT_INJECTION]: The skill ingests migration task parameters from user-provided Excel files.
  • Ingestion points: The excel_reader.py script reads host and proxy configuration from .xlsx files.
  • Boundary markers: While data is structured within the Excel sheets, there are no explicit delimiters or instructions to the AI to ignore potential instructions embedded within text fields.
  • Capability inventory: The skill has high-privilege capabilities including executing arbitrary commands via SSH and managing cloud infrastructure through the hcloud CLI.
  • Sanitization: The implementation includes regex-based validation for ECS names and task identifiers in task_name_utils.py and ownership_utils.py to prevent malicious payloads from reaching cloud APIs.
  • [EXTERNAL_DOWNLOADS]: The skill retrieves necessary binaries from established external sources, including the GOST proxy tool from its official GitHub repository, the migration agent from the vendor's object storage service, and rsync source code from Samba.org.
  • [CREDENTIALS_UNSAFE]: The skill manages Huawei Cloud Access Keys and Secret Keys. It implements a CredentialManager that utilizes an AES-256-GCM encrypted local cache, mandates user confirmation for credential updates, and masks sensitive identifiers in log outputs to prevent exposure.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Sep 30, 2026, 08:05 AM